ml4t-engineer

Feature engineering for financial machine learning: validated features, labeling methods, alternative bars, and leakage-safe dataset preparation.

Part of the ML4T Library Ecosystem

This library is one of six interconnected libraries supporting the machine learning for trading workflow described in Machine Learning for Trading:

Together they cover data infrastructure, feature engineering, modeling, signal evaluation, strategy backtesting, and live deployment.

What This Library Does

Transforming raw price data into predictive features is a core task in quantitative research. ml4t-engineer provides:

• 120 registry features across 11 categories (momentum, volatility, trend, microstructure, and more)
• Triple-barrier, ATR-based, percentile, trend-scanning, and meta-labeling methods from Advances in Financial Machine Learning
• Alternative bar sampling (volume bars, dollar bars, tick imbalance bars)
• Dataset building, preprocessing, and feature discovery for leakage-safe ML workflows

The library is built on Polars with Numba JIT compilation for numerical operations. 60 features are validated against TA-Lib at 1e-6 tolerance.

Installation

pip install ml4t-engineer

Optional dependencies:

pip install ml4t-engineer[ta]        # TA-Lib backend
pip install ml4t-engineer[viz]       # Visualization
pip install ml4t-engineer[calendars] # Trading calendars

Quick Start

import polars as pl
from ml4t.engineer import compute_features

df = pl.read_parquet("ohlcv.parquet")

# Compute features with default parameters
result = compute_features(df, ["rsi", "macd", "atr", "obv"])

# Or with custom parameters
result = compute_features(df, [
    {"name": "rsi", "params": {"period": 20}},
    {
        "name": "bollinger_bands",
        "params": {"period": 20, "nbdevup": 2.0, "nbdevdn": 2.0},
    },
])

Feature Registry

from ml4t.engineer.core.registry import get_registry

registry = get_registry()
print(registry.list_all())                    # All 120 features
print(registry.list_by_category("momentum"))  # 31 momentum indicators
print(registry.list_ta_lib_compatible())      # 60 TA-Lib validated features
print(registry.list_normalized())             # 37 bounded (0-100, -1 to 1)

Feature Categories

Triple-Barrier Labeling

from ml4t.engineer.config import LabelingConfig
from ml4t.engineer.labeling import triple_barrier_labels, atr_triple_barrier_labels

# Fixed barriers
tb_config = LabelingConfig.triple_barrier(
    upper_barrier=0.02,    # 2% profit target
    lower_barrier=0.01,    # 1% stop loss
    max_holding_period=20, # 20 bars
)
labels = triple_barrier_labels(
    df,
    config=tb_config,
)

# ATR-based dynamic barriers
atr_config = LabelingConfig.atr_barrier(
    atr_tp_multiple=2.0,
    atr_sl_multiple=1.0,
    atr_period=14,
    max_holding_period=20,
)
labels = atr_triple_barrier_labels(
    df,
    config=atr_config,
)

# Time-based horizons
tb_time_config = LabelingConfig.triple_barrier(
    upper_barrier=0.02,
    lower_barrier=0.01,
    max_holding_period="4h",  # 4 hours
)
labels = triple_barrier_labels(
    df,
    config=tb_time_config,
)

Alternative Bars

from ml4t.engineer.bars import VolumeBarSampler, DollarBarSampler, TickImbalanceBarSampler

# Volume bars (equal volume per bar)
vbars = VolumeBarSampler(volume_per_bar=1000).sample(tick_data)

# Dollar bars (equal dollar volume per bar)
dbars = DollarBarSampler(dollars_per_bar=1_000_000).sample(tick_data)

# Tick imbalance bars (information-driven)
ibars = TickImbalanceBarSampler(expected_ticks_per_bar=100).sample(tick_data)

Documentation

• Docs Home - library overview and workflow map
• Quickstart - first working feature and labeling workflow
• Features - 120 features across 11 registry categories
• Labeling - 7 labeling methods for supervised learning
• Dataset Builder - leakage-safe train/test preparation
• Examples - runnable scripts for complete workflows and focused features

Technical Characteristics

• Polars-native: All computations use Polars expressions
• Numba-accelerated: JIT compilation for numerical kernels
• TA-Lib validated: 60 features validated at 1e-6 tolerance
• AFML-compliant: Labeling methods verified against Advances in Financial Machine Learning
• ML-ready outputs: 37 features produce bounded outputs (0-100, -1 to 1) for direct model input; remaining features work with standard preprocessing (returns, z-scores, robust scaling)

Related Libraries

• ml4t-specs: Shared feed and artifact schema definitions across the ML4T stack
• ml4t-data: Market data acquisition and storage
• ml4t-diagnostic: Signal evaluation and statistical validation
• ml4t-backtest: Event-driven backtesting
• ml4t-live: Live trading with broker integration

Development

git clone https://github.com/ml4t/engineer.git
cd ml4t-engineer
uv sync
uv run pytest tests/ -q
uv run ty check

References

• Lopez de Prado, M. (2018). Advances in Financial Machine Learning. Wiley.
• Lopez de Prado, M. (2020). Machine Learning for Asset Managers. Cambridge.

License

MIT License - see LICENSE for details.